CN112219028B

CN112219028B - Wind turbine control method

Info

Publication number: CN112219028B
Application number: CN201980037657.6A
Authority: CN
Inventors: J·韦斯特高; A·S·尼尔森; J·P·穆尔西亚
Original assignee: Vestas Wind Systems AS
Current assignee: Vestas Wind Systems AS
Priority date: 2018-05-07
Filing date: 2019-05-02
Publication date: 2023-11-28
Anticipated expiration: 2039-05-02
Also published as: ES2966993T3; CN112219028A; WO2019214785A1; US20210123416A1; EP3791060A1; EP3791060B1; EP3791060C0; US11713747B2

Abstract

A method of operating a wind turbine. The wind turbine is operated during an operation period according to a control strategy. A sensor signal is received from a sensor that measures an operating parameter of the turbine during an operating period. The model is used to obtain a modeled fatigue value based on the sensor signal. The modeled fatigue values provide an estimate of fatigue loads applied to components of the turbine over an operating period. The control strategy is modified based on the modeled fatigue values.

Description

Wind turbine control method

Technical Field

The present invention relates to control of wind turbines, and more particularly to control of wind turbines based on fatigue load values.

Background

When addressing a wind turbine or a wind power plant, it is common to first measure the weather conditions of the proposed site. From these measurements, the fatigue load to which the turbine will be subjected can be predicted and a control strategy designed to operate the turbine over its lifetime. The designed control strategy may include static control features such as wind sector operation modes, which may, for example, set the turbine to operate in a given sector or shut down the turbine in a given sector. The control strategy designed may be designed to balance turbine life, operating expenses, and annual energy production. The turbine is then constructed and operated based on a control strategy for its predetermined lifetime (which may be between 20-25 years).

However, the predetermined control strategy for such a long period of time does not allow for actual field conditions during operation of the turbine in the general control strategy of the turbine.

It is against this background that the present invention has been devised.

Disclosure of Invention

In a first aspect, there is provided a method of operating a wind turbine, the method comprising:

operating the wind turbine for an operating period according to a control strategy;

receiving a sensor signal from a sensor measuring an operating parameter of the turbine during the operating period;

obtaining a modeled fatigue value based on the sensor signal using a model, wherein the modeled fatigue value provides an estimate of fatigue load applied to a component of the turbine during the period of operation; and

the control strategy is modified based on the modeled fatigue values.

By modeling the fatigue load based on actual operational measurements measured by the sensors, the control strategy may be optimized for the actual operational parameters experienced by the turbine. For example, control strategies in terms of power curves, thrust limits (possibly in a wind sector management scheme) may be adjusted.

The sensor may be a strain gauge, but strain gauges may be expensive. Thus, preferably, the sensor is not a strain gauge.

Optionally, the modeled fatigue value is obtained by: statistical data is derived from the sensor signals and input into the model to obtain the modeled fatigue value, thereby obtaining a modeled fatigue value. For example, the statistics may include one or more of an average, a maximum, a minimum, a standard deviation, a rain count statistic, a load duration/revolution distribution statistic of the operating parameter over the operating period. The operating parameters may exhibit periodic variations over the operating period. Deriving the statistical data may be more computationally efficient than recording and using the overall change in the operating parameter over time.

In some embodiments, the model may be a proxy model. The proxy model may include a response surface that predicts fatigue values when given a sensor signal (or a set of sensor signals, each signal related to a different operating parameter) as input. The use of proxy models may provide a computationally efficient method for predicting fatigue values.

The method may further include training the proxy model using training data. Training the model using the training data may include running a machine learning algorithm based on the training data.

In some such embodiments, the proxy model may be trained using simulated training data from the aeroelastic turbine load simulation. Aeroelastic load simulation is typically based on blade element momentum theory (BEM), which models the aerodynamic properties of the rotor and structural models of turbines and control systems. The simulated input may be a representation of the turbulence of the wind based on the turbulence model. The simulated output may be turbine response (e.g., blade deflection, rotational speed or acceleration, load, and power generation). Previously recorded statistics (e.g., statistics from other turbines or turbine farms) may be used as inputs to the simulation. The simulation may output a predicted fatigue value based on the input. The simulation may be validated using strain gauges on the prototype turbine. By using such simulations, a large amount of data may be generated to train the proxy model, thereby ensuring that the proxy model accurately estimates fatigue loads when used with actual measurements.

In some embodiments, the method may include determining whether the modeled fatigue value is different from an expected fatigue value; and modifying the control strategy based on the modeled fatigue value if the modeled fatigue value is different from the expected fatigue value.

For example, the method may include: estimating a fatigue life of the turbine prior to operating the turbine during said period of operation; determining an expected fatigue value for the operating period based on the fatigue life and a duration of the operating period; and comparing the modeled fatigue value with the expected fatigue value to determine whether the modeled fatigue value is different from the expected fatigue value. In this way, it may be determined whether the fatigue load experienced by the turbine due to the actual operating parameters is higher or lower than the expected fatigue load at that point in the turbine life.

In certain embodiments, modifying the control strategy may include: the control strategy is modified to reduce a difference between the modeled fatigue value and the expected fatigue value.

Fatigue usage measurements (such as usage) may be defined based on differences or ratios between the modeled fatigue values and the expected fatigue values.

In some embodiments, the operating period includes a plurality of operating intervals, and the modeled fatigue value is a cumulative modeled fatigue value obtained by: using the model to obtain an interval fatigue value for each operating interval based on the sensor signal for that interval; and accumulating the interval fatigue value for each operating interval. Thus, the modeled fatigue values over the operating period may represent cumulative modeled fatigue values over a plurality of operating intervals.

The operating parameters may include wind parameters (such as wind speed).

The operating parameters may include at least one of: wind speed; tower acceleration; rotor speed; blade pitch angle; active power and blade root bending moment. Measuring the value of such an operating parameter may be relatively inexpensive compared to directly measuring the fatigue load using a strain gauge. The sensor may measure one or more operating parameters; or multiple sensors may be used, each measuring an operating parameter.

The model may be used to obtain only a single modeled fatigue value that provides an estimate of fatigue loads applied to only a single component of the turbine during an operational period. Alternatively, the model may be used to obtain a plurality of modeled fatigue values, each of which provides an estimate of fatigue loads applied to various different components of the turbine over an operating period. For example, the model may be used to obtain modeled fatigue values for blade lag/edge direction moments, drive train torque, drive train bending moment, and/or tower bending moment.

Modifying the control strategy may include adjusting a control setting (such as a power setting or a speed setting) of the turbine based on the modeled fatigue value.

In particular embodiments, modifying the control strategy may include adjusting a pitch yaw control gain of the wind turbine. Adjusting the control gain may reduce blade pitching activity, thereby increasing the life of the blade bearing.

Modifying the control strategy may additionally or alternatively include adjusting at least one of: the life expectancy of the turbine; an operating mode of the turbine; a power setting of the turbine; and speed setting of the turbine.

According to a second aspect, there is provided a wind turbine control system adapted to perform the method according to any embodiment of the first aspect. The wind turbine control system may comprise a controller arranged to receive sensor signals from the sensors.

In general, a controller may be a unit or a collection of functional units that includes one or more processors, input/output interfaces, and a memory capable of storing instructions executable by the processors.

According to a third aspect, there is provided a wind turbine comprising a wind turbine control system according to the second aspect.

According to a fourth aspect, a computer program product is provided, comprising software code adapted to control a wind turbine when executed on a data processing system, the computer program product being adapted to perform the method according to any embodiment of the first aspect.

The computer program product may be provided on a computer readable storage medium or may be downloaded from a communication network. The computer program product includes instructions that, when loaded onto a data processing system (e.g., in the form of a controller), cause the data processing system to execute the instructions.

In general, the various aspects of the invention may be combined and coupled in any way possible within the scope of the invention. These and other aspects, features and/or advantages of the present invention will become apparent from and elucidated with reference to the embodiments described hereinafter.

Drawings

Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

FIG. 1 illustrates an example of a wind turbine;

FIG. 2 illustrates an embodiment of a control system and elements of a wind turbine;

FIG. 3a illustrates a set of training data that may be used to train a proxy model;

FIG. 3b shows a curve fit of the training data of FIG. 3 a;

FIG. 4 illustrates an example usage curve;

FIG. 5 illustrates a modified control strategy in which the life expectancy of the turbine is adjusted;

FIG. 6 illustrates a modified control strategy in which the control settings of the turbine are adjusted; and

FIG. 7 illustrates a method of controlling a wind turbine.

Detailed Description

Fig. 1 shows an example of a wind turbine 1 in a schematic perspective view. The wind turbine 1 comprises a tower 2, a nacelle 3 at the tower apex, and a rotor 4 operatively coupled to a generator housed inside the nacelle 3. In addition to the generator, the nacelle also houses the various components required to convert wind energy into electrical energy and the various components required to operate, control and optimize the performance of the wind turbine 1. The rotor 4 of the wind turbine comprises a central hub 5 and a plurality of blades 6 protruding outwards from the central hub 5. In the embodiment shown, the rotor 4 comprises three blades 6, but the number may vary. Furthermore, the wind turbine comprises a control system. The control system may be placed inside the nacelle or distributed at various locations inside the turbine and communicatively connected.

The wind turbine 1 may be comprised in a collection of other wind turbines belonging to a wind power plant (also called wind farm or wind farm) which is used as a power plant connected to the grid by means of a transmission line. The power grid is typically composed of a network of power stations, transmission circuits and substations, the substations being linked by a transmission line network, the transmission lines transmitting power to loads in the form of end users and other customers of the power company.

Fig. 2 schematically shows an embodiment of a control system 20 and elements of a wind turbine. The wind turbine comprises rotor blades 6, which rotor blades 6 are mechanically connected to a generator 22 via a gearbox 23. In direct drive systems and other systems, the gearbox 23 may not be present. The power generated by the generator 22 is injected into the grid 24 via an electrical converter 25. The generator 22 and the converter 25 may be based on a Full Scale Converter (FSC) architecture or a Doubly Fed Induction Generator (DFIG) architecture, although other types may be used.

The control system 20 includes a plurality of elements including at least one main controller 200 having a processor and memory, such that the processor is capable of performing computing tasks based on instructions stored in the memory. Typically, the wind turbine controller ensures that the wind turbine generates a required level of power output in operation. This is obtained by adjusting the pitch angle of the blades 6 and/or the power extraction of the converter 25. To this end, the control system comprises: a pitch system comprising a pitch controller 27 using a pitch reference 28; and a power system including a power controller 29 that uses the power reference 26. The wind turbine rotor includes rotor blades that may be pitched by a pitch mechanism. The rotor may comprise a common pitch system that adjusts all pitch angles of all rotor blades simultaneously, and furthermore the rotor comprises thereon a separate pitch system that enables separate pitching of the rotor blades. The control system or elements of the control system may be placed in a power plant controller (not shown) so that the turbine may be operated based on externally provided instructions.

Typically, the turbine operates according to a life control strategy that is determined before the turbine begins to operate. The lifetime control strategy is based on a model with known limitations that estimate the actual field conditions and on static control features such as wind sector operation modes.

The present invention provides a method of controlling a wind turbine, wherein during operation of the turbine, a control strategy is modified based on modeled actual fatigue values (i.e. fatigue values estimated based on actual measured data) experienced by the turbine. Such adaptive control of the turbine enables balancing of life, operational expenditure (OPEX) and Annual Energy Production (AEP) to achieve better turbine control relative to actual field conditions.

Modeling fatigue values

The proxy model 31 is used to obtain modeled fatigue values based on the sensor signals. The proxy model 31 receives as input measurement data from sensors 30 on or associated with the turbine. The sensor 30 measures one or more operating parameters experienced by the turbine. The operating parameters may include wind speed, tower acceleration, rotor speed, blade pitch angle, active power, and/or measured blade root bending moment.

The sensors 30 may be sensors that are part of a supervisory control and data acquisition ("SCADA") system. SCADA systems are large-scale distributed measurement and control systems that are commonly used to monitor or control chemical, physical or transportation processes, where the term SCADA generally refers to a central system that monitors and controls the entire site.

The proxy model 31 outputs one or more modeled fatigue values. Each modeled fatigue value provides an estimate of fatigue load applied to a component of the turbine over a period of time. The proxy model may provide an estimate of the fatigue load to be measured by strain gauges on components of the turbine. For example, the proxy model may provide estimates of fatigue loads caused by blade lag/edge direction moments, drive train torque, drive train bending moments (e.g., bending of the drive train in a vertical or horizontal direction), and/or tower bending moments. The output may be expressed as a rain count and/or a load duration and/or a revolution distribution. The output may be a fatigue value determined from such an estimated load value. For example, periodic variations in the estimated load value may be converted to a measure of fatigue load using the Palmgren-Miner linear impairment assumption (Miner rule).

The proxy model may be created as follows: a machine learning technique is used to construct a response surface that predicts fatigue values for a given sensor input. The proxy model response surface may be trained using training data created by the simulation tool. For example, a aeroelastic turbine load simulation tool may be used. As will be familiar to those skilled in the art, the simulation may utilize a generic software package such as Flex5 or HAWC 2. The aeroelastic simulation may be used to predict fatigue values that result from a widely varying set of operating parameters. As will be appreciated by those skilled in the art, these predicted fatigue values are used to train the proxy model by applying a machine learning algorithm.

Once the proxy model is created, it may be used to estimate or model the fatigue loads experienced by each turbine component based on the actual measurement data from the sensors 30.

To avoid the operational and computational costs required to fully record the changes in the operating parameters over the operating period, statistical data based on the measurements may be entered into the proxy model instead of raw sensor data. The statistical data may be a statistical representation of the operating parameters experienced by the turbine during the operating period. For example, an average, maximum, minimum, standard deviation, rain flow count, load duration, and/or load profile over the operating period may be determined and fed into the proxy model.

The fatigue value may be modeled for only a single operating period or for multiple operating periods (i.e., a series of operating intervals). The run time period may have a duration of between 5 minutes and 15 minutes, for example, or may be 10 minutes, or may be between 1 second and 10 seconds. Any other duration of the run period may be used. The modeled fatigue values may be obtained online (i.e., in real-time immediately after the end of the corresponding operating period). Alternatively, the modeled fatigue values may be obtained offline based on the recorded sensor data.

Fig. 3a and 3b give examples of how the proxy model is trained based on training data from the aeroelastic simulation. Figure 3a shows a set of training data. The x-axis of fig. 3a represents the values of a particular operating parameter (in this case the average wind speed V) during the simulated operating period.

The y-axis of fig. 3a represents fatigue values from training data determined by aeroelastic simulation. In this example, the fatigue value represents a fatigue tilt load at the main bearing of the wind turbine.

Each data point, indicated by a cross in fig. 3a, represents a fatigue value calculated by inputting a data set into the aeroelastic simulation, the data set representing a change in the operating parameter (in this case wind speed V) over the simulated operating period. The aeroelastic simulation then processes the data set to determine the pitch load at the main bearing, and then calculates the fatigue value for that period of operation using the Miner rule. The average wind speed V of the data set input into the aeroelastic simulation is then correlated with a particular fatigue value to generate data points in the training data set. For example, data point 30 in FIG. 3a shows about 1750 fatigue values associated with about 26 average wind speeds V, and data point 31 shows about 1500 fatigue values associated with about 13 average wind speeds V.

The training data represented in fig. 3a is then used to train the proxy model 31. Specifically, the machine learning algorithm analyzes the training data to identify correlations between the set of operating parameters and the fatigue values, which may then be used to predict fatigue loads when sensor data from the real wind turbine is input into the proxy model. In particular, the machine learning algorithm may identify a correlation between a particular operating parameter and a fatigue value. The proxy model may thus be trained using those operating parameters that are most closely related to fatigue values in the training data.

Using the example shown in fig. 3a, there is a correlation between the average wind speed V and the fatigue value, and a machine learning algorithm fits a curve 32 to the training data, as shown in fig. 3 b. In effect, this curve 32 is a one-dimensional proxy model that may be used to map any value of the average wind speed V to a modeled fatigue value, thereby providing an estimate of the fatigue load applied to the main bearings of the turbine during the operational period.

The process shown in fig. 3a and 3b may then be repeated based on another operating parameter to generate a two-dimensional response surface (two-dimensional equivalent of curve 32), and so on to generate a higher dimension.

In practice, many different sets of input and output variables will be simulated by aeroelastic simulation to train the proxy model. The proxy model uses all the data sets to form a multidimensional response surface instead of the simple curve 32 in the one-dimensional example of fig. 3 b.

Preferably, a set of multiple operating parameters measured by multiple sensors associated with the turbine are input into the proxy model. For example, a set of sensor signals representing two, three, five or more operating parameters may be used. Using more operating parameters allows more accurate modeled fatigue values to be obtained.

Comparing the modeled fatigue value with the estimated fatigue value

As described above, the modeled fatigue value may be compared to the expected fatigue value of the turbine at that point in the turbine's life. In particular embodiments, usage may be determined to aid in comparing the modeled fatigue value to an expected fatigue value. The usage may represent the cumulative fatigue load experienced by the turbine since the start of operation-i.e., it may represent the sum of modeled fatigue values for a plurality of operating intervals.

Fatigue value usage may be defined as the "actual" modeled fatigue value divided by the expected fatigue value at that point in the expected life of the turbine. The life expectancy may be the estimated life of the wind turbine before the wind turbine begins to operate based on an initial control strategy.

If x is the total operating time so far, the wind turbine has been operating within an operating period from time 0 to time x. The operation period is composed of a plurality of operation intervals. For example, if x is one day and the duration of each operating interval is 10 minutes, the operating period so far includes 144 operating intervals. The proxy model 31 outputs an interval fatigue value for each 10 minute interval based on the sensor signals within the interval. By accumulating the interval fatigue values for each operation period, the accumulated fatigue value, in other words, the operation sum of the interval fatigue values output by the retention agent model 31 can be obtained. Thus, in the above example, the accumulated modeled fatigue values are obtained by accumulating the previous 144 outputs of the proxy model.

The Usage (UR) can then be calculated as:

UR＝d _x /((x/T)*d _T )

where x is the total run time so far; t is the total expected (design) lifetime; d, d _x Is the cumulative modeling fatigue value at time x-i.e., the sum of the interval fatigue values output by the proxy model 31 during the run period from time 0 to time x; d _T Is the design fatigue value. The design fatigue value is the expected fatigue value at the end of life expectancy (time T) as predicted before the turbine begins to operate.

The usage may be calculated for the entire turbine, or individual usage may be defined for one or more components of the turbine, such as usage indicative of cumulative blade lag/edge direction moment, drive train torque, drive train bending moment (pitch/yaw) and/or tower bending moment. In the case of individual usage, d in the UR equation _x Is the cumulative modeled fatigue value for only that particular component/load at time x; d, d _T Is the design fatigue value for that particular component/load only.

Fig. 4 shows a component UR of an example turbine after an operating period x. Line 40 in fig. 4 represents a UR curve that is expected based on the design fatigue value and life expectancy of the turbine (as predicted before the turbine begins to operate). The expected UR-curve 40 reaches a value of 1 at the end of the design life T, which indicates that the turbine has been subjected to all the fatigue loads it is designed to withstand.

Lines 42-44 represent component UR curves over an operational period up to time x based on modeled fatigue values output by proxy model 31. In this example, each of the lines 42-44 is located below the expected UR curve 40, indicating that each component is subjected to a lower than expected fatigue load. Thus, the turbine has the potential to be unused for withstanding fatigue loads, which provides an opportunity to modify the control strategy as described in more detail below.

Line 42 represents the component of the wind turbine that has heretofore had the greatest modeling usage. The line represents a component with minimal potential for bearing additional fatigue loads and thus may be particularly important for modifying the control strategy.

On the other hand, if any of the component usage rates is above the expected UR curve 40, it is indicated that the component is subjected to fatigue loads that exceed the expected value at that point in the turbine life. In this case, modifying the control strategy may attempt to reduce at least future fatigue loading of the component.

While the usage rate may be particularly useful for comparing the modeled fatigue value to the expected fatigue value, other comparison measures may be used. For example, a simple difference between the modeled and expected fatigue values may be calculated and the control strategy modified based on the difference.

Modifying control strategy

By comparing the modeled fatigue value experienced by the turbine (e.g., using usage) to the expected fatigue value at that point in the turbine's life, the unused potential for experiencing fatigue loading can be identified. The overall control strategy of the turbine may be modified to take advantage of this unused potential. This may ensure that the turbine is optimally used throughout its lifetime, for example to extract maximum energy while minimizing operational expenditure.

One possible modification is to adjust the working life of the turbine. Typically, the life of the turbine is estimated before operation begins, and a life control strategy is determined. However, if the modeled fatigue values indicate that there is potential to withstand additional fatigue loads, the lifetime may be extended. For example, in the case where the usage has been calculated and plotted in FIG. 4, the modeled UR curve may be extrapolated in time until UR reaches 1, as shown by extrapolated UR curve 42a in FIG. 5. The operating time t+Δ when the extrapolated UR-curve 42a reaches 1 may be considered a new estimated life of the turbine. For example, the remaining runtime can be calculated as: remaining time (e.g., in years) =t/UR-x. The life extending operation may be performed for the entire turbine UR or for the largest component UR (i.e. line 42 in fig. 4).

Modifying the life of the turbine may include calculating a new life expectancy of the turbine based on the modeled fatigue values, and updating a record of the life expectancy of the turbine with the new life expectancy. The record of life expectancy may be stored on an external storage device, such as a storage device associated with a remote network. For example, the external storage device may store a database that associates a plurality of turbines with respective life expectancy values. The turbine (or the turbine's controller) may communicate with an external storage device via a wired or wireless communication link to update the life expectancy value of the turbine in a database.

Additionally or alternatively, the operation of particular components may be regulated. For example, pitch response control gain may be adjusted based on estimated component fatigue usage. The pitch response control gain may be related to a so-called pitch-yaw control (TYC), which control gain is also referred to as TYC gain. The purpose of pitch yaw control is to reduce the out-of-plane loads (so-called pitch and yaw loads) experienced by the main bearings by introducing periodic pitching (sometimes also referred to as independent pitching) which counteracts the out-of-plane loads (and to some extent reduces the blade fatigue values) on the main bearings experienced by the turbine. Periodic pitching of the TYC control results in increased wear of the blade bearings due to increased pitching activity. If the UR-curve affected by the TYC control is lower than expected, the TYC gain (amplitude of the periodical pitch) can be reduced. This results in a reduced pitch activity and thus an increased lifetime of the blade bearing. This in turn reduces the operational expenditure costs of operating the turbine.

The adjustment of the TYC gain may be constrained by the desired UR-curve (i.e. line 40 in fig. 4) -i.e. the TYC gain may be adjusted such that the associated modeled pitch/yaw UR-curve approximates the desired UR-curve. Alternatively, if the pitch/yaw UR curve is not the highest of the modeled UR curves (i.e., not line 42 in fig. 4), the TYC gain may be adjusted until the pitch/yaw UR is the highest of the modeled UR curves.

Modifying the control strategy may additionally or alternatively comprise adjusting an operational setting of the turbine (e.g. a wind sector operational mode) based on the modeled fatigue value or the corresponding usage. In particular, this may include adjusting the power and speed settings of the turbine.

Typically, turbines have different available modes of operation (i.e., modes with alternative power and speed settings). The motivation for having these modes is to be able to tailor the operation of the turbine to a specific site, for example to take advantage of wind sectors with benign conditions by operating the turbine in a more aggressive mode (i.e. high power and/or speed) or to reduce the fatigue load of the turbine in wind sectors with severe conditions (i.e. low speed and/or power). These modes are typically selected based on site-specific load predictions prior to construction, and the turbine is typically operated at these settings throughout its life. However, in the present invention, these settings can be modified if the estimated UR-curve shows unused potential for additional fatigue loads. In particular, in the event that there is unused potential, a more aggressive mode may be selected for future operation. Conversely, if it is determined that the turbine is experiencing a greater fatigue load than expected, a less aggressive mode may be selected.

Modifying the control strategy may additionally or alternatively include adjusting the damping control settings based on modeled tower fatigue values or corresponding usage rates. The damping control setting may be a control setting that affects the damping provided by the turbine or a component of the turbine. For example, the damping control setting may be pitch control. The damping may be lateral tower damping (SSTD) and/or front-rear tower damping (FATD). The damping control may be a pitch response gain value or an activation threshold defining a condition in which pitch control damping is enabled. Thus, if an estimated UR-curve (e.g. tower load UR-curve) affected by SSTD or FATD is found to be lower than the expected UR-curve, the pitching activity for reducing the tower load may be reduced. This may increase the lifetime of the blade bearing, thereby reducing the OPEX cost.

Fig. 6 gives an example in which the control strategy associated with UR-curve 42 is modified to become more aggressive at time x, so it is expected that future usage will follow curve 42b—reaching usage 1 at time T.

Any combination of the above control strategies may be used to optimize the life, OPEX and annual energy production of the turbine.

FIG. 7 illustrates a method of controlling a wind turbine according to an embodiment of the invention.

In step 51, the wind turbine is operated for a first operation period according to the control strategy. In particular, the control strategy may be an initial control strategy that is determined before the turbine begins to operate.

In step 52, one or more sensor signals are received from sensors measuring one or more of the operating parameters of the turbine during a first operating period.

In step 53, fatigue values applied to components of the turbine during a first period of operation are modeled based on the sensor signals. In particular, when provided with sensor signals as input, a proxy model may be used to estimate fatigue values.

In step 54, a control strategy of the turbine is modified based on the modeled fatigue values.

The method may then begin again, returning to step 51 to operate the wind turbine for a second period of operation according to the modified control strategy. After the second period of operation, the method may proceed to steps 52 through 54 to further modify the control strategy based on the modeled fatigue values for the second period of operation.

The steps of fig. 7 may be implemented as a computer program product or code adapted to generate instructions to a controller arranged to control the operation of the wind turbine or a component of the wind turbine. The computer program may be provided in any suitable manner. The computer program product is typically stored and executed by the wind turbine control system 200.

Although the present invention has been described in connection with the specified embodiments, it should not be construed as being limited in any way to the examples presented. The invention may be implemented by any suitable means; and the scope of the invention is to be construed in accordance with the appended claims. Any reference signs in the claims shall not be construed as limiting the scope.

Claims

1. A method of operating a wind turbine, the method comprising:

operating a wind turbine for an operational period according to an initial control strategy, wherein the initial control strategy comprises a first usage curve over the lifetime of the wind turbine;

receiving sensor signals from sensors measuring operational parameters of the wind turbine during the operational period;

obtaining a modeled fatigue value based on the sensor signal using a model, wherein the modeled fatigue value provides an estimate of a fatigue load applied to a component of the wind turbine during the period of operation, wherein the model comprises a proxy model trained using training data, and wherein the modeled fatigue value comprises at least one correlation between an operating parameter of the component and the estimate of the fatigue load applied to the component based on the training data;

determining a second usage curve over the operating period using the modeled fatigue values; and

modifying the initial control strategy based on a difference between the first and second usage curves during the operational period to generate a modified control strategy comprising a third usage curve for a future operational period, wherein the third usage curve intersects the first usage curve at a future time value.

2. The method of claim 1, wherein the modeled fatigue value is obtained by: statistical data is derived from the sensor signals and input into the model to obtain the modeled fatigue values.

3. The method of claim 1, wherein the proxy model is trained using training data from a aeroelastic turbine load simulation.

4. The method according to claim 1, wherein the method comprises: determining whether the modeled fatigue value is different from an expected fatigue value; and modifying the initial control strategy based on the modeled fatigue value if the modeled fatigue value is different from the expected fatigue value.

5. The method of claim 4, wherein the method further comprises:

estimating a fatigue life of the wind turbine prior to operating the wind turbine during the operating period;

determining an expected fatigue value for the operating period based on the fatigue life and a duration of the operating period; and

the modeled fatigue value is compared to the expected fatigue value to determine whether the modeled fatigue value is different from the expected fatigue value.

6. The method of claim 1, further comprising: defining a fatigue usage measurement based on a difference or ratio between the modeled fatigue value and an expected fatigue value; and modifying the initial control strategy based on the fatigue usage measurements.

7. The method of claim 1, wherein modifying the initial control strategy comprises: the initial control strategy is modified to reduce the difference between the modeled fatigue value and an expected fatigue value.

8. The method according to claim 1, wherein:

the operating period includes a plurality of operating intervals; and is also provided with

The modeled fatigue value is a cumulative modeled fatigue value, the cumulative modeled fatigue value obtained by: using the model to obtain an interval fatigue value for each operating interval based on the sensor signal for that interval; and accumulating the interval fatigue value for each operating interval.

9. The method of claim 1, wherein the operating parameters include at least one of: wind speed; tower acceleration; rotor speed; pitch angle; active power and blade root bending moment.

10. The method of claim 1, wherein the operating parameter comprises a wind parameter.

11. The method of claim 1, wherein the modeled fatigue value is for a blade lag/edge direction moment, a drive train torque, a drive train bending moment, or a tower bending moment.

12. The method of claim 1, wherein modifying the initial control strategy comprises: adjusting a control setting of the wind turbine based on the modeled fatigue value.

13. The method of claim 1, wherein modifying the initial control strategy comprises: adjusting a pitch response gain of the wind turbine.

14. A wind turbine comprising a control system adapted to perform the method according to any preceding claim.

15. A computer program product comprising software code adapted to control a wind turbine when executed on a data processing system, the computer program product being adapted to perform the method according to any one of claims 1 to 13.